docs: describe the advisory planning benchmark

This commit is contained in:
nessar committed 2026-10-10 16:42:04 +02:00
1 parent 98e87fecee
commit ae68e31575
1 file changed
+3 -1
+3 -1
View File
@@ -247,7 +247,9 @@ GitLab CI runs independent validation, test, build, security, and image stages:
Pipelines switch to MR pipelines when a branch has an open MR, avoiding duplicate builds. Docker publication from MR pipelines is limited to detached pipelines within this project. Tag pipelines promote the already-built commit image instead of rebuilding it.
The test job runs Vitest directly with one worker and a 2 GiB Node heap limit. It runs application tests with coverage, then the complete benchmark protocol separately without coverage instrumentation. The benchmark uses a five-minute execution timeout to allow for shared-runner contention; all protocol assertions remain mandatory. Timing measurements are benchmark evidence, not CI performance thresholds.
The required `test` job runs application tests with coverage, one worker and a 2 GiB Node heap limit. A separate automatic `benchmark` job uses the same environment outside coverage instrumentation, with the unchanged five-minute timeout and protocol assertions.
At the user's request, the benchmark job has `allow_failure: true` while #78 is open, so benchmark failure does not block functional delivery. #78 must restore the required benchmark gate after the protocol passes. A green pipeline during this exception does not establish benchmark acceptance.
### GitFlow and releases